Quip Among Us

A creative word-smithing party game for 3-8 players. A recreation of the Jackbox game Quiplash, with a twist: one of the answers each round is generated by AI. Adapted from quiplash-js.

Playable in the same room or remotely via screensharing.

How to Play

• Host starts a new game at https://quip-among-us-9549d28ae7de.herokuapp.com/create
  • If players are remote, screen share this screen
• Players join game at https://einarbalan.com/quipamongus (recommended to use a mobile device)
• Host starts the game
• Players receive two prompts to answer (Be as silly as possible)
• Players then vote for their favorite answer to different prompts, as well as which answer they think is generated by AI
• Whoever gets the most votes for their submitted answer is technically the winner, but the person that had the most fun is the true winner

Tech Stack

• Frontend - React
• Backend - Nodejs, Express
• SocketIO

Extracting answers for the prompts

The Python script(state/generated-data/extract.py) helps generate funny responses for a collection of prompts stored in state/generated-data/Prompts.pg13.js using the Gemini 1.5 Flash AI model. The script processes prompts and instructs Gemini to respond with witty, humorous answers, saving both prompts and responses to a CSV file for easy tracking and review.

How to Start Local Server

• npm start

Note: The React web app needs to get built, whereas the node server does not.

Testing on another device

• Start local server
• If you have a firewall, ensure it allows connections from NodeJS
• Open your other device's browser and connect to HOST_IP:3001

Evaluation

• ./Evaluation - Evaluation with BertScore
• CS263_Evaluation.ipynb
  • A Jupyter Notebook for running and visualizing evaluation processes interactively.
  • Use this file to explore the evaluation pipeline step by step and verify results.
• evaluate_quiplash.py
  • Automates the evaluation of JSON files.
  • Saves results in the ./results/ directory.

How to Run Evaluation

1. Place your JSON files in the ./reports/ directory.
2. Run the evaluation script:

   python evaluate_quiplash.py

3. If you want to compare the BertScore by files and model, run the CS263_Evaluation.ipynb file.

Directory Structure

• build - Built web app
• public - Template html from create-react-app
• server - Node Server code
• src - Front-end code
• Evaluation - Evaluation with BertScore